Characterizing position-specific properties of siRNAs

RNA Interference (RNAi) is a new technique in biology and medical research field that can silence one or more genes of interest. One crucial step in RNAi experiments is designing short interfering RNA (siRNA) sequences that are both effective in mediating robust knockdown as well as exhibiting high specificity. In this study, we used logistic regression to weight siRNA sequences from a siRNA database, siRecords, where the siRNA efficacy was measured in categorical values. The model built with logistic regression in our current study demonstrated several advantages over the one built in a previous study on the same dataset, siRecords. First, our approach is multi-variable in nature, i.e., we examine all the features as a whole instead of one feature at a time. Second, our model uses fewer features than the previous study while maintaining equal prediction accuracy with that of Support Vector Machines (SVMs). Third, our model uses numerical weights to indicate the importance of each feature regarding siRNA effectiveness, while the previous study uses a predefined threshold such as a P value of 0.05, to measure the statistical significance of the features. Finding the right threshold for the P value is often data dependent and adhoc. Fourth, our features are consistent, i.e., features that can boost the 90% siRNA efficacy can also boost the 70% siRNA efficacy, a property that is missing in the model of the previous study. Copyright